Application of genetic algorithm in selection of dominant input variables in sensor fault diagnosis of nonlinear systems

Industrial processes rely heavily on information provided by sensors. Reliability of sensor data is vital to assure an acceptable performance of these complex and nonlinear processes. In this paper, the analytical redundancy approach has been adopted to detect and isolate sensor faults in which the model of a given nonlinear dynamical system is identified based on the available input/output time profile. Towards this goal, an evolving Takagi-Sugeno approach as a universal approximator is used to represent a nonlinear mapping between the past values of input/output data and the current value of the output data. However, the main challenge is the selection of the appropriate set of past values that can lead to the best estimate of the output. In this paper, a genetic algorithm is utilized as a powerful data-driven tool for finding the best set of input-output past values. The proposed approach is applied to the problem of sensor fault detection and isolation in a Continuous-Flow Stirred-Tank Reactor. Simulation results demonstrate and validate the performance capabilities of the proposed approach.